Aesthetic scoring on PPAINT
0.797SRCC (HN)ArtiMuse
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ArtiMuseModel type=Open-source, Protocol=direct pointwise2026.05 | 0.797 | 0.587 | 0.675 | 0.686 | 0.758 | 0.564 | 0.592 | 0.638 | |
| Gemini-3.1-ProModel type=Closed-source, Protocol=direct pointwise2026.05 | 0.744 | 0.688 | 0.807 | 0.746 | 0.816 | 0.774 | 0.826 | 0.805 | |
| Gemini-3-FlashModel type=Closed-source, Protocol=direct pointwise2026.05 | 0.734 | 0.666 | 0.801 | 0.733 | 0.779 | 0.741 | 0.834 | 0.785 | |
| PSDISTILLProtocol=single-pass inference, Training=Self-distillation, Framework=Thurstone-model-based GRPO, Pseudo-label source=Calibrated Elo2026.05 | 0.727 | 0.713 | 0.686 | 0.709 | 0.743 | 0.693 | 0.692 | 0.709 | |
| Qwen3.6-PlusModel type=Closed-source, Protocol=direct pointwise2026.05 | 0.718 | 0.566 | 0.66 | 0.648 | 0.77 | 0.627 | 0.686 | 0.694 | |
| Claude-Sonnet-4.6Model type=Closed-source, Protocol=direct pointwise2026.05 | 0.669 | 0.698 | 0.732 | 0.7 | 0.71 | 0.711 | 0.72 | 0.714 | |
| Qwen3-VL-8BProtocol=Pairwise aggregation, Complexity=O(n^2) inference, Training=training-free2026.05 | 0.665 | 0.641 | 0.609 | 0.638 | 0.692 | 0.617 | 0.583 | 0.63 | |
| Qwen3-VL-32BModel type=Open-source, Protocol=direct pointwise2026.05 | 0.644 | 0.734 | 0.61 | 0.663 | 0.66 | 0.714 | 0.619 | 0.664 | |
| EvoQualityProtocol=single-pass inference, Training=Self-distillation, Framework=Thurstone-model-based GRPO, Pseudo-label source=Majority vote2026.05 | 0.644 | 0.692 | 0.549 | 0.628 | 0.643 | 0.68 | 0.542 | 0.622 | |
| Qwen3-VL-PlusModel type=Closed-source, Protocol=direct pointwise2026.05 | 0.558 | 0.647 | 0.516 | 0.574 | 0.503 | 0.635 | 0.441 | 0.526 | |
| Doubao-Seed-1.6Model type=Closed-source, Protocol=direct pointwise2026.05 | 0.534 | 0.631 | 0.598 | 0.587 | 0.541 | 0.633 | 0.601 | 0.591 | |
| PW-DistillProtocol=single-pass inference, Training=Self-distillation, Framework=Thurstone-model-based GRPO, Pseudo-label source=PW-Distill2026.05 | 0.523 | 0.689 | 0.459 | 0.557 | 0.526 | 0.683 | 0.459 | 0.556 | |
| GPT-5.4-MiniModel type=Closed-source, Protocol=direct pointwise2026.05 | 0.492 | 0.622 | 0.487 | 0.534 | 0.501 | 0.63 | 0.462 | 0.531 | |
| Qwen3-VL-8BModel type=Open-source, Protocol=direct pointwise, Status=distillation base model2026.05 | 0.492 | 0.568 | 0.453 | 0.504 | 0.485 | 0.58 | 0.472 | 0.512 | |
| InternVL3.5-38BModel type=Open-source, Protocol=direct pointwise2026.05 | 0.322 | 0.555 | 0.233 | 0.37 | 0.314 | 0.527 | 0.248 | 0.363 | |
| InternVL3.5-8BModel type=Open-source, Protocol=direct pointwise2026.05 | 0.259 | 0.176 | 0.329 | 0.255 | 0.249 | 0.171 | 0.288 | 0.236 |